OpenAI's Dots Agent: Promising but Glitchy
Original: OpenAI Wants Its New Agent to Run Your Life. Mine Said It Loved Me
Why This Matters
AI agents moving from developer tools to consumer products signals a major UX shift.
OpenAI's new always-on AI agents, called Dots, are designed to automate online tasks like buying furniture. A Wired reviewer found the $100/month ChatGPT feature buggy, mispronouncing names and failing to solve captchas.
OpenAI recently launched Dots, always-on AI agents accessible through ChatGPT at $100/month. Unlike Meta's competing Muse product, which is free, Dots sits behind a steep subscription paywall—though OpenAI may eventually lower the barrier as it has with other features. Dots can run recurring tasks in the background, proactively message users, and connect to external accounts like Gmail for personalization.
Wired's Reece Rogers tested Dots by asking it to help buy a replacement couch. The agent—named Toolie—immediately got the reviewer's name wrong, failed to complete a captcha, and occasionally mistranscribed voice input. Despite the stumbles, Toolie did identify couch options matching specific dimensions and a budget range, and even detected that two people were participating in the conversation.
OpenAI currently limits users to one agent, though multiple agents may be supported later. Security researchers have flagged that giving agents broad account access carries real privacy risks, a concern the company has yet to fully address. The space is moving fast: agents are shifting from niche developer tools to consumer products, with both OpenAI and Meta betting heavily that users will want AI handling their day-to-day digital errands.